Utilizing Causal Learning for Cognitive Management of 6G Networks

In this paper, we study the use of causal reasoning for a cognitive network where we investigate causal discovery and its use in the context of 6G. The fundamental requirement in causal inference is to discover the causal relation among variables in a complex system (such as among a number of network KPIs). Once such causal relations are correctly captured, the cognitive capability of a network can significantly be increased by having the ability of causal and counterfactual reasoning. We envision that this can additionally bolster the trust and transparency to such cognitive system. First, we attempt to discover a causal graph that represents the causal relation in a networking setup. By applying the state-of-the-art causal discovery algorithms that include both constraint and score-based methods, we aim at uncovering which network KPIs and actions are causally impacting each other. It is observed that the causal discovery algorithms are able to capture the causal relations between the network KPIs and actions, reasonably well. Then, we give a use-case example within an autonomous network context, and evaluate the estimated causal graph with our realistic network emulator to predict the impact of an intervention on a network KPI. The preliminary results are promising and pave the way for more focused future studies that can support complex commercial networks in a scalable way.